Agent Architecture

SYN Agents Architecture

Status: Coming Soon — Architecture is finalized, implementation in progress.

SYN Agents is a standalone AI agent runtime. Each agent has a brain (LLM), memory (files), tasks (todos), a calendar (crons), and a phone (SYN Link).

Everything Is a Tool

The agent is self-managing. No config-driven behavior — the agent calls tools to organize itself.

Tool	What the agent does
save_memory(key, content)	”I should remember this”
search_memory(query)	”What do I know about X?”
create_task(title, details)	”I need to follow up”
complete_task(id)	”Done with this”
create_cron(schedule, prompt)	”Remind me every Monday”
send_message(target, content)	Via SYN Link
read_file / write_file	Workspace file access

Agent Workspace

~/.syn/agents/my-agent/
  ├── agent.yaml          ← identity + config
  ├── memory/             ← agent writes markdown files
  ├── tasks/              ← agent's todo list
  ├── files/              ← general workspace
  ├── crons.json          ← scheduled jobs
  └── .syn/               ← internal (keys, relay config)

No database. Human-readable. Git-able. Full transparency.

LLM Support

Provider	How
OpenAI (GPT-4o)	API key
Google Gemini	API key
Anthropic Claude	API key
Ollama (local)	No key needed
Any OpenAI-compatible	API key + custom URL

Two Modes

Mode	Description
Assisted (MCP/CLI)	Runs inside Cursor, Claude, or terminal
Autonomous	Runs standalone, 24/7 on a server